ia64/linux-2.6.18-xen.hg

view drivers/net/3c501.c @ 897:329ea0ccb344

balloon: try harder to balloon up under memory pressure.

Currently if the balloon driver is unable to increase the guest's
reservation it assumes the failure was due to reaching its full
allocation, gives up on the ballooning operation and records the limit
it reached as the "hard limit". The driver will not try again until
the target is set again (even to the same value).

However it is possible that ballooning has in fact failed due to
memory pressure in the host and therefore it is desirable to keep
attempting to reach the target in case memory becomes available. The
most likely scenario is that some guests are ballooning down while
others are ballooning up and therefore there is temporary memory
pressure while things stabilise. You would not expect a well behaved
toolstack to ask a domain to balloon to more than its allocation nor
would you expect it to deliberately over-commit memory by setting
balloon targets which exceed the total host memory.

This patch drops the concept of a hard limit and causes the balloon
driver to retry increasing the reservation on a timer in the same
manner as when decreasing the reservation.

Also if we partially succeed in increasing the reservation
(i.e. receive less pages than we asked for) then we may as well keep
those pages rather than returning them to Xen.

Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
author Keir Fraser <keir.fraser@citrix.com>
date Fri Jun 05 14:01:20 2009 +0100 (2009-06-05)
parents 831230e53067
children
line source
1 /* 3c501.c: A 3Com 3c501 Ethernet driver for Linux. */
2 /*
3 Written 1992,1993,1994 Donald Becker
5 Copyright 1993 United States Government as represented by the
6 Director, National Security Agency. This software may be used and
7 distributed according to the terms of the GNU General Public License,
8 incorporated herein by reference.
10 This is a device driver for the 3Com Etherlink 3c501.
11 Do not purchase this card, even as a joke. It's performance is horrible,
12 and it breaks in many ways.
14 The original author may be reached as becker@scyld.com, or C/O
15 Scyld Computing Corporation
16 410 Severn Ave., Suite 210
17 Annapolis MD 21403
19 Fixed (again!) the missing interrupt locking on TX/RX shifting.
20 Alan Cox <Alan.Cox@linux.org>
22 Removed calls to init_etherdev since they are no longer needed, and
23 cleaned up modularization just a bit. The driver still allows only
24 the default address for cards when loaded as a module, but that's
25 really less braindead than anyone using a 3c501 board. :)
26 19950208 (invid@msen.com)
28 Added traps for interrupts hitting the window as we clear and TX load
29 the board. Now getting 150K/second FTP with a 3c501 card. Still playing
30 with a TX-TX optimisation to see if we can touch 180-200K/second as seems
31 theoretically maximum.
32 19950402 Alan Cox <Alan.Cox@linux.org>
34 Cleaned up for 2.3.x because we broke SMP now.
35 20000208 Alan Cox <alan@redhat.com>
37 Check up pass for 2.5. Nothing significant changed
38 20021009 Alan Cox <alan@redhat.com>
40 Fixed zero fill corner case
41 20030104 Alan Cox <alan@redhat.com>
44 For the avoidance of doubt the "preferred form" of this code is one which
45 is in an open non patent encumbered format. Where cryptographic key signing
46 forms part of the process of creating an executable the information
47 including keys needed to generate an equivalently functional executable
48 are deemed to be part of the source code.
50 */
53 /**
54 * DOC: 3c501 Card Notes
55 *
56 * Some notes on this thing if you have to hack it. [Alan]
57 *
58 * Some documentation is available from 3Com. Due to the boards age
59 * standard responses when you ask for this will range from 'be serious'
60 * to 'give it to a museum'. The documentation is incomplete and mostly
61 * of historical interest anyway.
62 *
63 * The basic system is a single buffer which can be used to receive or
64 * transmit a packet. A third command mode exists when you are setting
65 * things up.
66 *
67 * If it's transmitting it's not receiving and vice versa. In fact the
68 * time to get the board back into useful state after an operation is
69 * quite large.
70 *
71 * The driver works by keeping the board in receive mode waiting for a
72 * packet to arrive. When one arrives it is copied out of the buffer
73 * and delivered to the kernel. The card is reloaded and off we go.
74 *
75 * When transmitting lp->txing is set and the card is reset (from
76 * receive mode) [possibly losing a packet just received] to command
77 * mode. A packet is loaded and transmit mode triggered. The interrupt
78 * handler runs different code for transmit interrupts and can handle
79 * returning to receive mode or retransmissions (yes you have to help
80 * out with those too).
81 *
82 * DOC: Problems
83 *
84 * There are a wide variety of undocumented error returns from the card
85 * and you basically have to kick the board and pray if they turn up. Most
86 * only occur under extreme load or if you do something the board doesn't
87 * like (eg touching a register at the wrong time).
88 *
89 * The driver is less efficient than it could be. It switches through
90 * receive mode even if more transmits are queued. If this worries you buy
91 * a real Ethernet card.
92 *
93 * The combination of slow receive restart and no real multicast
94 * filter makes the board unusable with a kernel compiled for IP
95 * multicasting in a real multicast environment. That's down to the board,
96 * but even with no multicast programs running a multicast IP kernel is
97 * in group 224.0.0.1 and you will therefore be listening to all multicasts.
98 * One nv conference running over that Ethernet and you can give up.
99 *
100 */
102 #define DRV_NAME "3c501"
103 #define DRV_VERSION "2002/10/09"
106 static const char version[] =
107 DRV_NAME ".c: " DRV_VERSION " Alan Cox (alan@redhat.com).\n";
109 /*
110 * Braindamage remaining:
111 * The 3c501 board.
112 */
114 #include <linux/module.h>
116 #include <linux/kernel.h>
117 #include <linux/fcntl.h>
118 #include <linux/ioport.h>
119 #include <linux/interrupt.h>
120 #include <linux/slab.h>
121 #include <linux/string.h>
122 #include <linux/errno.h>
123 #include <linux/config.h> /* for CONFIG_IP_MULTICAST */
124 #include <linux/spinlock.h>
125 #include <linux/ethtool.h>
126 #include <linux/delay.h>
127 #include <linux/bitops.h>
129 #include <asm/uaccess.h>
130 #include <asm/io.h>
132 #include <linux/netdevice.h>
133 #include <linux/etherdevice.h>
134 #include <linux/skbuff.h>
135 #include <linux/init.h>
137 #include "3c501.h"
139 /*
140 * The boilerplate probe code.
141 */
143 static int io=0x280;
144 static int irq=5;
145 static int mem_start;
147 /**
148 * el1_probe: - probe for a 3c501
149 * @dev: The device structure passed in to probe.
150 *
151 * This can be called from two places. The network layer will probe using
152 * a device structure passed in with the probe information completed. For a
153 * modular driver we use #init_module to fill in our own structure and probe
154 * for it.
155 *
156 * Returns 0 on success. ENXIO if asked not to probe and ENODEV if asked to
157 * probe and failing to find anything.
158 */
160 struct net_device * __init el1_probe(int unit)
161 {
162 struct net_device *dev = alloc_etherdev(sizeof(struct net_local));
163 static unsigned ports[] = { 0x280, 0x300, 0};
164 unsigned *port;
165 int err = 0;
167 if (!dev)
168 return ERR_PTR(-ENOMEM);
170 if (unit >= 0) {
171 sprintf(dev->name, "eth%d", unit);
172 netdev_boot_setup_check(dev);
173 io = dev->base_addr;
174 irq = dev->irq;
175 mem_start = dev->mem_start & 7;
176 }
178 SET_MODULE_OWNER(dev);
180 if (io > 0x1ff) { /* Check a single specified location. */
181 err = el1_probe1(dev, io);
182 } else if (io != 0) {
183 err = -ENXIO; /* Don't probe at all. */
184 } else {
185 for (port = ports; *port && el1_probe1(dev, *port); port++)
186 ;
187 if (!*port)
188 err = -ENODEV;
189 }
190 if (err)
191 goto out;
192 err = register_netdev(dev);
193 if (err)
194 goto out1;
195 return dev;
196 out1:
197 release_region(dev->base_addr, EL1_IO_EXTENT);
198 out:
199 free_netdev(dev);
200 return ERR_PTR(err);
201 }
203 /**
204 * el1_probe1:
205 * @dev: The device structure to use
206 * @ioaddr: An I/O address to probe at.
207 *
208 * The actual probe. This is iterated over by #el1_probe in order to
209 * check all the applicable device locations.
210 *
211 * Returns 0 for a success, in which case the device is activated,
212 * EAGAIN if the IRQ is in use by another driver, and ENODEV if the
213 * board cannot be found.
214 */
216 static int __init el1_probe1(struct net_device *dev, int ioaddr)
217 {
218 struct net_local *lp;
219 const char *mname; /* Vendor name */
220 unsigned char station_addr[6];
221 int autoirq = 0;
222 int i;
224 /*
225 * Reserve I/O resource for exclusive use by this driver
226 */
228 if (!request_region(ioaddr, EL1_IO_EXTENT, DRV_NAME))
229 return -ENODEV;
231 /*
232 * Read the station address PROM data from the special port.
233 */
235 for (i = 0; i < 6; i++)
236 {
237 outw(i, ioaddr + EL1_DATAPTR);
238 station_addr[i] = inb(ioaddr + EL1_SAPROM);
239 }
240 /*
241 * Check the first three octets of the S.A. for 3Com's prefix, or
242 * for the Sager NP943 prefix.
243 */
245 if (station_addr[0] == 0x02 && station_addr[1] == 0x60
246 && station_addr[2] == 0x8c)
247 {
248 mname = "3c501";
249 } else if (station_addr[0] == 0x00 && station_addr[1] == 0x80
250 && station_addr[2] == 0xC8)
251 {
252 mname = "NP943";
253 }
254 else {
255 release_region(ioaddr, EL1_IO_EXTENT);
256 return -ENODEV;
257 }
259 /*
260 * We auto-IRQ by shutting off the interrupt line and letting it float
261 * high.
262 */
264 dev->irq = irq;
266 if (dev->irq < 2)
267 {
268 unsigned long irq_mask;
270 irq_mask = probe_irq_on();
271 inb(RX_STATUS); /* Clear pending interrupts. */
272 inb(TX_STATUS);
273 outb(AX_LOOP + 1, AX_CMD);
275 outb(0x00, AX_CMD);
277 mdelay(20);
278 autoirq = probe_irq_off(irq_mask);
280 if (autoirq == 0)
281 {
282 printk(KERN_WARNING "%s probe at %#x failed to detect IRQ line.\n",
283 mname, ioaddr);
284 release_region(ioaddr, EL1_IO_EXTENT);
285 return -EAGAIN;
286 }
287 }
289 outb(AX_RESET+AX_LOOP, AX_CMD); /* Loopback mode. */
290 dev->base_addr = ioaddr;
291 memcpy(dev->dev_addr, station_addr, ETH_ALEN);
293 if (mem_start & 0xf)
294 el_debug = mem_start & 0x7;
295 if (autoirq)
296 dev->irq = autoirq;
298 printk(KERN_INFO "%s: %s EtherLink at %#lx, using %sIRQ %d.\n", dev->name, mname, dev->base_addr,
299 autoirq ? "auto":"assigned ", dev->irq);
301 #ifdef CONFIG_IP_MULTICAST
302 printk(KERN_WARNING "WARNING: Use of the 3c501 in a multicast kernel is NOT recommended.\n");
303 #endif
305 if (el_debug)
306 printk(KERN_DEBUG "%s", version);
308 memset(dev->priv, 0, sizeof(struct net_local));
309 lp = netdev_priv(dev);
310 spin_lock_init(&lp->lock);
312 /*
313 * The EL1-specific entries in the device structure.
314 */
316 dev->open = &el_open;
317 dev->hard_start_xmit = &el_start_xmit;
318 dev->tx_timeout = &el_timeout;
319 dev->watchdog_timeo = HZ;
320 dev->stop = &el1_close;
321 dev->get_stats = &el1_get_stats;
322 dev->set_multicast_list = &set_multicast_list;
323 dev->ethtool_ops = &netdev_ethtool_ops;
324 return 0;
325 }
327 /**
328 * el1_open:
329 * @dev: device that is being opened
330 *
331 * When an ifconfig is issued which changes the device flags to include
332 * IFF_UP this function is called. It is only called when the change
333 * occurs, not when the interface remains up. #el1_close will be called
334 * when it goes down.
335 *
336 * Returns 0 for a successful open, or -EAGAIN if someone has run off
337 * with our interrupt line.
338 */
340 static int el_open(struct net_device *dev)
341 {
342 int retval;
343 int ioaddr = dev->base_addr;
344 struct net_local *lp = netdev_priv(dev);
345 unsigned long flags;
347 if (el_debug > 2)
348 printk(KERN_DEBUG "%s: Doing el_open()...", dev->name);
350 if ((retval = request_irq(dev->irq, &el_interrupt, 0, dev->name, dev)))
351 return retval;
353 spin_lock_irqsave(&lp->lock, flags);
354 el_reset(dev);
355 spin_unlock_irqrestore(&lp->lock, flags);
357 lp->txing = 0; /* Board in RX mode */
358 outb(AX_RX, AX_CMD); /* Aux control, irq and receive enabled */
359 netif_start_queue(dev);
360 return 0;
361 }
363 /**
364 * el_timeout:
365 * @dev: The 3c501 card that has timed out
366 *
367 * Attempt to restart the board. This is basically a mixture of extreme
368 * violence and prayer
369 *
370 */
372 static void el_timeout(struct net_device *dev)
373 {
374 struct net_local *lp = netdev_priv(dev);
375 int ioaddr = dev->base_addr;
377 if (el_debug)
378 printk (KERN_DEBUG "%s: transmit timed out, txsr %#2x axsr=%02x rxsr=%02x.\n",
379 dev->name, inb(TX_STATUS), inb(AX_STATUS), inb(RX_STATUS));
380 lp->stats.tx_errors++;
381 outb(TX_NORM, TX_CMD);
382 outb(RX_NORM, RX_CMD);
383 outb(AX_OFF, AX_CMD); /* Just trigger a false interrupt. */
384 outb(AX_RX, AX_CMD); /* Aux control, irq and receive enabled */
385 lp->txing = 0; /* Ripped back in to RX */
386 netif_wake_queue(dev);
387 }
390 /**
391 * el_start_xmit:
392 * @skb: The packet that is queued to be sent
393 * @dev: The 3c501 card we want to throw it down
394 *
395 * Attempt to send a packet to a 3c501 card. There are some interesting
396 * catches here because the 3c501 is an extremely old and therefore
397 * stupid piece of technology.
398 *
399 * If we are handling an interrupt on the other CPU we cannot load a packet
400 * as we may still be attempting to retrieve the last RX packet buffer.
401 *
402 * When a transmit times out we dump the card into control mode and just
403 * start again. It happens enough that it isnt worth logging.
404 *
405 * We avoid holding the spin locks when doing the packet load to the board.
406 * The device is very slow, and its DMA mode is even slower. If we held the
407 * lock while loading 1500 bytes onto the controller we would drop a lot of
408 * serial port characters. This requires we do extra locking, but we have
409 * no real choice.
410 */
412 static int el_start_xmit(struct sk_buff *skb, struct net_device *dev)
413 {
414 struct net_local *lp = netdev_priv(dev);
415 int ioaddr = dev->base_addr;
416 unsigned long flags;
418 /*
419 * Avoid incoming interrupts between us flipping txing and flipping
420 * mode as the driver assumes txing is a faithful indicator of card
421 * state
422 */
424 spin_lock_irqsave(&lp->lock, flags);
426 /*
427 * Avoid timer-based retransmission conflicts.
428 */
430 netif_stop_queue(dev);
432 do
433 {
434 int len = skb->len;
435 int pad = 0;
436 int gp_start;
437 unsigned char *buf = skb->data;
439 if (len < ETH_ZLEN)
440 pad = ETH_ZLEN - len;
442 gp_start = 0x800 - ( len + pad );
444 lp->tx_pkt_start = gp_start;
445 lp->collisions = 0;
447 lp->stats.tx_bytes += skb->len;
449 /*
450 * Command mode with status cleared should [in theory]
451 * mean no more interrupts can be pending on the card.
452 */
454 outb_p(AX_SYS, AX_CMD);
455 inb_p(RX_STATUS);
456 inb_p(TX_STATUS);
458 lp->loading = 1;
459 lp->txing = 1;
461 /*
462 * Turn interrupts back on while we spend a pleasant afternoon
463 * loading bytes into the board
464 */
466 spin_unlock_irqrestore(&lp->lock, flags);
468 outw(0x00, RX_BUF_CLR); /* Set rx packet area to 0. */
469 outw(gp_start, GP_LOW); /* aim - packet will be loaded into buffer start */
470 outsb(DATAPORT,buf,len); /* load buffer (usual thing each byte increments the pointer) */
471 if (pad) {
472 while(pad--) /* Zero fill buffer tail */
473 outb(0, DATAPORT);
474 }
475 outw(gp_start, GP_LOW); /* the board reuses the same register */
477 if(lp->loading != 2)
478 {
479 outb(AX_XMIT, AX_CMD); /* fire ... Trigger xmit. */
480 lp->loading=0;
481 dev->trans_start = jiffies;
482 if (el_debug > 2)
483 printk(KERN_DEBUG " queued xmit.\n");
484 dev_kfree_skb (skb);
485 return 0;
486 }
487 /* A receive upset our load, despite our best efforts */
488 if(el_debug>2)
489 printk(KERN_DEBUG "%s: burped during tx load.\n", dev->name);
490 spin_lock_irqsave(&lp->lock, flags);
491 }
492 while(1);
494 }
496 /**
497 * el_interrupt:
498 * @irq: Interrupt number
499 * @dev_id: The 3c501 that burped
500 * @regs: Register data (surplus to our requirements)
501 *
502 * Handle the ether interface interrupts. The 3c501 needs a lot more
503 * hand holding than most cards. In particular we get a transmit interrupt
504 * with a collision error because the board firmware isnt capable of rewinding
505 * its own transmit buffer pointers. It can however count to 16 for us.
506 *
507 * On the receive side the card is also very dumb. It has no buffering to
508 * speak of. We simply pull the packet out of its PIO buffer (which is slow)
509 * and queue it for the kernel. Then we reset the card for the next packet.
510 *
511 * We sometimes get surprise interrupts late both because the SMP IRQ delivery
512 * is message passing and because the card sometimes seems to deliver late. I
513 * think if it is part way through a receive and the mode is changed it carries
514 * on receiving and sends us an interrupt. We have to band aid all these cases
515 * to get a sensible 150kBytes/second performance. Even then you want a small
516 * TCP window.
517 */
519 static irqreturn_t el_interrupt(int irq, void *dev_id, struct pt_regs *regs)
520 {
521 struct net_device *dev = dev_id;
522 struct net_local *lp;
523 int ioaddr;
524 int axsr; /* Aux. status reg. */
526 ioaddr = dev->base_addr;
527 lp = netdev_priv(dev);
529 spin_lock(&lp->lock);
531 /*
532 * What happened ?
533 */
535 axsr = inb(AX_STATUS);
537 /*
538 * Log it
539 */
541 if (el_debug > 3)
542 printk(KERN_DEBUG "%s: el_interrupt() aux=%#02x", dev->name, axsr);
544 if(lp->loading==1 && !lp->txing)
545 printk(KERN_WARNING "%s: Inconsistent state loading while not in tx\n",
546 dev->name);
548 if (lp->txing)
549 {
551 /*
552 * Board in transmit mode. May be loading. If we are
553 * loading we shouldn't have got this.
554 */
556 int txsr = inb(TX_STATUS);
558 if(lp->loading==1)
559 {
560 if(el_debug > 2)
561 {
562 printk(KERN_DEBUG "%s: Interrupt while loading [", dev->name);
563 printk(KERN_DEBUG " txsr=%02x gp=%04x rp=%04x]\n", txsr, inw(GP_LOW),inw(RX_LOW));
564 }
565 lp->loading=2; /* Force a reload */
566 spin_unlock(&lp->lock);
567 goto out;
568 }
570 if (el_debug > 6)
571 printk(KERN_DEBUG " txsr=%02x gp=%04x rp=%04x", txsr, inw(GP_LOW),inw(RX_LOW));
573 if ((axsr & 0x80) && (txsr & TX_READY) == 0)
574 {
575 /*
576 * FIXME: is there a logic to whether to keep on trying or
577 * reset immediately ?
578 */
579 if(el_debug>1)
580 printk(KERN_DEBUG "%s: Unusual interrupt during Tx, txsr=%02x axsr=%02x"
581 " gp=%03x rp=%03x.\n", dev->name, txsr, axsr,
582 inw(ioaddr + EL1_DATAPTR), inw(ioaddr + EL1_RXPTR));
583 lp->txing = 0;
584 netif_wake_queue(dev);
585 }
586 else if (txsr & TX_16COLLISIONS)
587 {
588 /*
589 * Timed out
590 */
591 if (el_debug)
592 printk (KERN_DEBUG "%s: Transmit failed 16 times, Ethernet jammed?\n",dev->name);
593 outb(AX_SYS, AX_CMD);
594 lp->txing = 0;
595 lp->stats.tx_aborted_errors++;
596 netif_wake_queue(dev);
597 }
598 else if (txsr & TX_COLLISION)
599 {
600 /*
601 * Retrigger xmit.
602 */
604 if (el_debug > 6)
605 printk(KERN_DEBUG " retransmitting after a collision.\n");
606 /*
607 * Poor little chip can't reset its own start pointer
608 */
610 outb(AX_SYS, AX_CMD);
611 outw(lp->tx_pkt_start, GP_LOW);
612 outb(AX_XMIT, AX_CMD);
613 lp->stats.collisions++;
614 spin_unlock(&lp->lock);
615 goto out;
616 }
617 else
618 {
619 /*
620 * It worked.. we will now fall through and receive
621 */
622 lp->stats.tx_packets++;
623 if (el_debug > 6)
624 printk(KERN_DEBUG " Tx succeeded %s\n",
625 (txsr & TX_RDY) ? "." : "but tx is busy!");
626 /*
627 * This is safe the interrupt is atomic WRT itself.
628 */
630 lp->txing = 0;
631 netif_wake_queue(dev); /* In case more to transmit */
632 }
633 }
634 else
635 {
636 /*
637 * In receive mode.
638 */
640 int rxsr = inb(RX_STATUS);
641 if (el_debug > 5)
642 printk(KERN_DEBUG " rxsr=%02x txsr=%02x rp=%04x", rxsr, inb(TX_STATUS),inw(RX_LOW));
643 /*
644 * Just reading rx_status fixes most errors.
645 */
646 if (rxsr & RX_MISSED)
647 lp->stats.rx_missed_errors++;
648 else if (rxsr & RX_RUNT)
649 { /* Handled to avoid board lock-up. */
650 lp->stats.rx_length_errors++;
651 if (el_debug > 5)
652 printk(KERN_DEBUG " runt.\n");
653 }
654 else if (rxsr & RX_GOOD)
655 {
656 /*
657 * Receive worked.
658 */
659 el_receive(dev);
660 }
661 else
662 {
663 /*
664 * Nothing? Something is broken!
665 */
666 if (el_debug > 2)
667 printk(KERN_DEBUG "%s: No packet seen, rxsr=%02x **resetting 3c501***\n",
668 dev->name, rxsr);
669 el_reset(dev);
670 }
671 if (el_debug > 3)
672 printk(KERN_DEBUG ".\n");
673 }
675 /*
676 * Move into receive mode
677 */
679 outb(AX_RX, AX_CMD);
680 outw(0x00, RX_BUF_CLR);
681 inb(RX_STATUS); /* Be certain that interrupts are cleared. */
682 inb(TX_STATUS);
683 spin_unlock(&lp->lock);
684 out:
685 return IRQ_HANDLED;
686 }
689 /**
690 * el_receive:
691 * @dev: Device to pull the packets from
692 *
693 * We have a good packet. Well, not really "good", just mostly not broken.
694 * We must check everything to see if it is good. In particular we occasionally
695 * get wild packet sizes from the card. If the packet seems sane we PIO it
696 * off the card and queue it for the protocol layers.
697 */
699 static void el_receive(struct net_device *dev)
700 {
701 struct net_local *lp = netdev_priv(dev);
702 int ioaddr = dev->base_addr;
703 int pkt_len;
704 struct sk_buff *skb;
706 pkt_len = inw(RX_LOW);
708 if (el_debug > 4)
709 printk(KERN_DEBUG " el_receive %d.\n", pkt_len);
711 if ((pkt_len < 60) || (pkt_len > 1536))
712 {
713 if (el_debug)
714 printk(KERN_DEBUG "%s: bogus packet, length=%d\n", dev->name, pkt_len);
715 lp->stats.rx_over_errors++;
716 return;
717 }
719 /*
720 * Command mode so we can empty the buffer
721 */
723 outb(AX_SYS, AX_CMD);
724 skb = dev_alloc_skb(pkt_len+2);
726 /*
727 * Start of frame
728 */
730 outw(0x00, GP_LOW);
731 if (skb == NULL)
732 {
733 printk(KERN_INFO "%s: Memory squeeze, dropping packet.\n", dev->name);
734 lp->stats.rx_dropped++;
735 return;
736 }
737 else
738 {
739 skb_reserve(skb,2); /* Force 16 byte alignment */
740 skb->dev = dev;
741 /*
742 * The read increments through the bytes. The interrupt
743 * handler will fix the pointer when it returns to
744 * receive mode.
745 */
746 insb(DATAPORT, skb_put(skb,pkt_len), pkt_len);
747 skb->protocol=eth_type_trans(skb,dev);
748 netif_rx(skb);
749 dev->last_rx = jiffies;
750 lp->stats.rx_packets++;
751 lp->stats.rx_bytes+=pkt_len;
752 }
753 return;
754 }
756 /**
757 * el_reset: Reset a 3c501 card
758 * @dev: The 3c501 card about to get zapped
759 *
760 * Even resetting a 3c501 isnt simple. When you activate reset it loses all
761 * its configuration. You must hold the lock when doing this. The function
762 * cannot take the lock itself as it is callable from the irq handler.
763 */
765 static void el_reset(struct net_device *dev)
766 {
767 struct net_local *lp = netdev_priv(dev);
768 int ioaddr = dev->base_addr;
770 if (el_debug> 2)
771 printk(KERN_INFO "3c501 reset...");
772 outb(AX_RESET, AX_CMD); /* Reset the chip */
773 outb(AX_LOOP, AX_CMD); /* Aux control, irq and loopback enabled */
774 {
775 int i;
776 for (i = 0; i < 6; i++) /* Set the station address. */
777 outb(dev->dev_addr[i], ioaddr + i);
778 }
780 outw(0, RX_BUF_CLR); /* Set rx packet area to 0. */
781 outb(TX_NORM, TX_CMD); /* tx irq on done, collision */
782 outb(RX_NORM, RX_CMD); /* Set Rx commands. */
783 inb(RX_STATUS); /* Clear status. */
784 inb(TX_STATUS);
785 lp->txing = 0;
786 }
788 /**
789 * el1_close:
790 * @dev: 3c501 card to shut down
791 *
792 * Close a 3c501 card. The IFF_UP flag has been cleared by the user via
793 * the SIOCSIFFLAGS ioctl. We stop any further transmissions being queued,
794 * and then disable the interrupts. Finally we reset the chip. The effects
795 * of the rest will be cleaned up by #el1_open. Always returns 0 indicating
796 * a success.
797 */
799 static int el1_close(struct net_device *dev)
800 {
801 int ioaddr = dev->base_addr;
803 if (el_debug > 2)
804 printk(KERN_INFO "%s: Shutting down Ethernet card at %#x.\n", dev->name, ioaddr);
806 netif_stop_queue(dev);
808 /*
809 * Free and disable the IRQ.
810 */
812 free_irq(dev->irq, dev);
813 outb(AX_RESET, AX_CMD); /* Reset the chip */
815 return 0;
816 }
818 /**
819 * el1_get_stats:
820 * @dev: The card to get the statistics for
821 *
822 * In smarter devices this function is needed to pull statistics off the
823 * board itself. The 3c501 has no hardware statistics. We maintain them all
824 * so they are by definition always up to date.
825 *
826 * Returns the statistics for the card from the card private data
827 */
829 static struct net_device_stats *el1_get_stats(struct net_device *dev)
830 {
831 struct net_local *lp = netdev_priv(dev);
832 return &lp->stats;
833 }
835 /**
836 * set_multicast_list:
837 * @dev: The device to adjust
838 *
839 * Set or clear the multicast filter for this adaptor to use the best-effort
840 * filtering supported. The 3c501 supports only three modes of filtering.
841 * It always receives broadcasts and packets for itself. You can choose to
842 * optionally receive all packets, or all multicast packets on top of this.
843 */
845 static void set_multicast_list(struct net_device *dev)
846 {
847 int ioaddr = dev->base_addr;
849 if(dev->flags&IFF_PROMISC)
850 {
851 outb(RX_PROM, RX_CMD);
852 inb(RX_STATUS);
853 }
854 else if (dev->mc_list || dev->flags&IFF_ALLMULTI)
855 {
856 outb(RX_MULT, RX_CMD); /* Multicast or all multicast is the same */
857 inb(RX_STATUS); /* Clear status. */
858 }
859 else
860 {
861 outb(RX_NORM, RX_CMD);
862 inb(RX_STATUS);
863 }
864 }
867 static void netdev_get_drvinfo(struct net_device *dev,
868 struct ethtool_drvinfo *info)
869 {
870 strcpy(info->driver, DRV_NAME);
871 strcpy(info->version, DRV_VERSION);
872 sprintf(info->bus_info, "ISA 0x%lx", dev->base_addr);
873 }
875 static u32 netdev_get_msglevel(struct net_device *dev)
876 {
877 return debug;
878 }
880 static void netdev_set_msglevel(struct net_device *dev, u32 level)
881 {
882 debug = level;
883 }
885 static struct ethtool_ops netdev_ethtool_ops = {
886 .get_drvinfo = netdev_get_drvinfo,
887 .get_msglevel = netdev_get_msglevel,
888 .set_msglevel = netdev_set_msglevel,
889 };
891 #ifdef MODULE
893 static struct net_device *dev_3c501;
895 module_param(io, int, 0);
896 module_param(irq, int, 0);
897 MODULE_PARM_DESC(io, "EtherLink I/O base address");
898 MODULE_PARM_DESC(irq, "EtherLink IRQ number");
900 /**
901 * init_module:
902 *
903 * When the driver is loaded as a module this function is called. We fake up
904 * a device structure with the base I/O and interrupt set as if it were being
905 * called from Space.c. This minimises the extra code that would otherwise
906 * be required.
907 *
908 * Returns 0 for success or -EIO if a card is not found. Returning an error
909 * here also causes the module to be unloaded
910 */
912 int __init init_module(void)
913 {
914 dev_3c501 = el1_probe(-1);
915 if (IS_ERR(dev_3c501))
916 return PTR_ERR(dev_3c501);
917 return 0;
918 }
920 /**
921 * cleanup_module:
922 *
923 * The module is being unloaded. We unhook our network device from the system
924 * and then free up the resources we took when the card was found.
925 */
927 void cleanup_module(void)
928 {
929 struct net_device *dev = dev_3c501;
930 unregister_netdev(dev);
931 release_region(dev->base_addr, EL1_IO_EXTENT);
932 free_netdev(dev);
933 }
935 #endif /* MODULE */
937 MODULE_AUTHOR("Donald Becker, Alan Cox");
938 MODULE_DESCRIPTION("Support for the ancient 3Com 3c501 ethernet card");
939 MODULE_LICENSE("GPL");